AI Watermarking

Cautions and Marking.

With AI Watermarking, Creators Strike Back Backdoor attacks regulate unauthorized uses of copyrighted or restricted data      By TAMMY XU

AI models rely on immense datasets to train their complex algorithms, but sometimes the use of those datasets for training purposes can infringe on the rights of the data owners. Yet actually proving that a model used a dataset without authorization has been notoriously difficult. However, a new studypublished in IEEE Transactions on Information Forensics and Security, researchers introduce a method for protecting datasets from unauthorized use by embedding digital watermarks into them. The technique could give data owners more say in who is allowed to train AI models using their data.

The simplest way of protecting datasets is to restrict their use, such as with encryption. But doing so would make those datasets difficult to use for authorized users as well. Instead, the researchers focused on detecting whether a given AI model was trained using a particular dataset, says the study’s lead author, Yiming Li. Models known to have been impermissibly trained on a dataset can be flagged for follow up by the data owner.

Watermarking methods could cause harm, too, though. Malicious actors, for instance, could teach a self-driving system to incorrectly recognize stop signs as speed limit signs.

The technique can be applied to many different types of machine learning problems, Li said, although the study focuses on classification models, including image classification. First, a small sample of images is selected from a dataset and a watermark consisting of a set pattern of altered pixels is embedded into each image. Then the classification label of each watermarked image is changed to correspond to a target label. This establishes a relationship between the watermark and the target label, creating what’s called a backdoor attack. Finally, the altered images are recombined with the rest of the dataset and published, where it’s available for consumption by both authorized users. To verify whether a particular model was trained using the dataset, researchers simply run watermarked images through the model and see whether they get back the target label.  .... ' 

3D Printing Poses a "Grave and Growing Threat" to Privacy

Had never thought of the premise suggested here, but yes is a consideration.   Grave?  Also mentions watermarking which we also studied for retail applications.

3D Printing Poses a "Grave and Growing Threat" to Privacy, Experts Warn
University of Exeter
September 8, 2020

Researchers at Durham University and the University of Exeter in the U.K. warn that three-dimensional (3D) printing technology poses a "grave and growing threat" to individual privacy and that governments and companies are unaware of these privacy issues. Said Exeter's James Griffin, "Every physical product that is 3D-printed has the potential to be tracked in a way that has never occurred before." The study is based on 30 in-depth interviews with representatives of Chinese 3D printing companies, most of whom believed the tracking technology incorporated into 3D printing would be used to handle piracy or copyright issues, and not for invading users’ privacy. The researchers called for a voluntary code of conduct that would encourage self-regulation of 3D printing and watermarking, and a specific software component that can isolate and protect private information collected from a watermark.

Watermarking Control Data for Safety from Hackers

Thoughtful and apparently simple idea.

Approach Could Protect Control Systems From Hackers
IEEE Spectrum
Michelle Hampson
March 26, 2020

Researchers at Siemens and Croatia’s University of Zagreb have developed a technique to more easily identify attacks against industrial control systems (ICS), like those used in the electric power grid, or to control traffic. The researchers applied the concept of "watermarking" data during transmission to ICS, in a manner that is broadly applicable without requiring details about the specific ICS. In such a scenario, when data is transmitted in real time over an unencrypted channel, it is accompanied by a specialized algorithm in the form of a recursive watermark (RWM) signal; any disruption to the RWM signal indicates an attack is underway. Said Siemens' Zhen Song, “If attackers change or delay the real-time channel signal a little bit, the algorithm can detect the suspicious event and raise alarms immediately."